There have been numerous attempts to provide a standardized definition ofa computational agent, but little consensus has emerged. We propose a simple test for agenthood that can be applied to a putative computational agent. Roughly, this test seeks to capture the intuition that an agent is an entity that can function as part ofa multiagent system. The test depends on the observed behavior of the supposed agent and not on the internals of it. We apply the test to some well-known kinds of systems of (supposed) agents and discuss the results. We present a formulation of the test and some variants with a semantics based on sociability. Our treatment of agenthood can thus serve as a methodological basis for evaluating putative agents and agent toolkits.